International Business Machines Corporation

Nuance Communications, Inc.

Stephen Mingyu Chu

Min Gong

Dong Sheng Li

Junchi Yan

Ning Duan

Jun Chi Yan

Xuan Zhang

Wei Peng Zhang

Weipeng Zhang

Vaibhava Goel

Etienne Marcheret

Chang Li

Gerasimos Potamianos

Bhuvana Ramabhadran

Peng Gao

Jing Chang Huang

Guo Qiang Hu

Liang Wu

Paul Joseph Vozila

Stanley S Chen

Ruhi Sarikaya

Yun Jie Qiu

Maximilian Bisani

Boliang Chen

Pankaj Srivastava

Yi Su

Tomomi Yoshioka

Tobias Enders

Junichi Yan

A computer-implemented method applies labels to unlabeled public data for use by a global model. One or more processors train one or more local machine learning models with local private data to create one or more trained models. Processor(s) generate a label for each of the local private data using the one or more trained models, where each label describes the local private data, and then apply the label to unlabeled public data to create labeled public data. One or more processors then input the labeled public data into a global model that uses the public data.

Training a machine learning model in a distributed privacy-preserving environment

A method, system, and computer program storage product determine determining a trajectory information type of a receipt submitted by an employee. Trajectory information associated with the receipt submitted by the employee is retrieved based on the trajectory information type. Trajectory information corresponding to a device associated with the employee is also retrieved. The receipt is determined as a valid receipt in response to the trajectory information associated with the receipt submitted by the employee matching the trajectory information associated with the device associated with the employee.

Expense compliance checking based on trajectory detection

Embodiments of the present disclosure provide method, system and computer program product for privacy preserving smart metering. According to one embodiment of the present disclosure, a plurality of customer-encrypted meter readings are received, each of the plurality of customer-encrypted meter readings being encrypted by a customer using a customer key, then a summation of customer keys is obtained, next the plurality of customer-encrypted meter readings are summed, and the summation of the plurality of customer-encrypted meter readings is decrypted based on the obtained summation of customer keys.

Privacy-preserving smart metering

Techniques for neural network training are provided. One computer-implemented method comprises: obtaining, by an electronic device operatively coupled to a processing unit, based on a similarity between two images, a first image and a second image. The computer-implemented method also comprises training, by the electronic device, a neural network based on the first image and the second image such that a distance between a first vector and a second vector generated respectively from the first image and the second image in the trained neural network is associated with the similarity.

Neural network training

Techniques for model based data processing are proposed. In one example, a computer-implemented method is as follows. A first model is determined from a first category of data processing models and a second model is determined from a second category of data processing models. The second category can be different from the first category in some instances. Performance of a first combined model of the first and second models can be compared with performance of the first model. A target model for processing data can be selected from the first combined model and the first model based on the comparing.

Model based data processing

Techniques are provided for providing a local computation service on client device authorized by a remote service provider while keeping local data private from the remote service provider. In one example, a computer-implemented method comprises transmitting, by a device operatively coupled to a processor, a homomorphic encryption public key and homomorphically encrypted input data to a service provider. The computer-implemented method also comprises receiving, by the device from the service provider, a homomorphically encrypted signature generated based on the homomorphic encryption public key, the homomorphically encrypted input data, and a homomorphic encryption equivalent defined function. The computer-implemented method also comprises in response to a determination, based on a defined function corresponding to the homomorphic encryption equivalent defined function, that the homomorphically encrypted signature and input data corresponding to the homomorphically encrypted input data are valid for a computation service, performing, by the device, the computation service on the input data.

Providing computation services with privacy

A method for determining an object referenced within a set of one or more informal online communications comprises: generating a knowledge graph for a company based at least on formal online communications, the knowledge graph comprising a plurality of node elements, and the knowledge graph further comprising, for each node element of the knowledge graph, a corresponding halo comprising one or more words which are temporally proximate to that node element within the formal online communications; for each node element of the knowledge graph which is determined to be present in a given informal online communication, detecting a halo comprising one or more words which are temporally proximate to that node element within the given informal online communication; and identifying which of the plurality of node elements has a corresponding halo within the knowledge graph most similar to the detected halo, wherein the identified node element is the referenced object.

Determining an object referenced within informal online communications

Embodiments of the present disclosure relate to a new approach for image orientation detection. In the computer-implemented method, at least one character area in an image is sampled. The orientation of the sampled character area is determined, and the orientation of the image is determined based on the determined orientation of the sampled character area.

Image orientation detection

Embodiments include a system, method, and computer program product for foreign organization name matching. Aspects include receiving a first entity name from a first database configured in a first language and receiving a second database, wherein the second database includes a plurality of entity names in a second language, wherein the first and second languages are different. Aspects also include performing an Internet search based on the first entity name, wherein a language setting of the Internet search is configured to the second language and identifying a list of candidate names based on a set of results from the Internet search. Another aspect includes filtering the list to obtain a resulting candidate name and determining whether the resulting candidate name matches one of the entity names in the second database. Aspects include associating the first entity name and a matching entity name in the second database based on the determination.

Foreign organization name matching

A language processing application employs a classing function optimized for the underlying production application context for which it is expected to process speech. A combination of class based and word based features generates a classing function optimized for a particular production application, meaning that a language model employing the classing function uses word classes having a high likelihood of accurately predicting word sequences encountered by a language model invoked by the production application. The classing function optimizes word classes by aligning the objective of word classing with the underlying language processing task to be performed by the production application. The classing function is optimized to correspond to usage in the production application context using class-based and word-based features by computing a likelihood of a word in an n-gram and a frequency of a word within a class of the n-gram.

Word classing for language modeling

A system and method for speech recognition includes determining active Gaussians related to a first feature stream and a second feature stream by labeling at least one of the first and second streams, and determining active Gaussians co-occurring in the first stream and the second stream based upon joint probability. A number of Gaussians computed is reduced based upon Gaussians already computed for the first stream and a number of Gaussians co-occurring in the second stream. Speech is decoded based on the Gaussians computed for the first and second streams.

Growing community of inventors

Beavercreek, OH, United States of America

Stephen Mingyu Chu